Question 1 (50 marks)

An important immigration shock studied in the labour economics literature is the Mariel Boatlift. On April 20, 1980, Fidel Castro announced that he would open the ports of Mariel, in Cuba, to anyone who would like to leave the country. As a result, many Cubans left for the United States. Most settled in Miami, increasing the labour force there by about 7% between May and September 1980. This question asks you to investigate the effect of this large immigration shock on Miami’s labour market.

Download the dataset marielper1.dta from the course website. The dataset contains information on the employment status, weekly wage and individual characteristics of individuals in Miami and other cities in the period from 1973 to 1991. The survey is conducted in May each year (you can assume that the effects of the immigration inflow would only potentially be seen in the survey from 1981 onwards).

a) Suppose that you believe that Atlanta, Los Angeles, Houston and Tampa-St Petersburg have similar characteristics to Miami and can be used as comparison cities. Complete the table below with the average (in percentage) of the unemployment indicator variable in Miami and the comparison cities in 1979 and 1981 and the differences between cities and across time. What are the difference-in-differences (DD) estimates of the effect of the immigration inflow on unemployment for each ethnic group? Use a simple regression to calculate the standard errors of the DD estimates. Are they statistically significant? Comment on the results.

Note: report the averages in percentage, rounded to three decimal points.

b) Repeat the analysis in part (a) for log weekly earnings, by completing the table below with the average log weekly earnings in Miami and the comparison cities in 1979 and 1981 and the differences between cities and across time. What are the difference- indifferences (DD) estimates of the effect of the immigration inflow on earnings for each ethnic group? Use a simple regression to calculate the standard errors of the DD estimates. Are they statistically significant? Comment on the results.

Note: report the averages rounded to three decimal points.

c) Instead of focusing on averages, you can make full use of the individual-level data for all years and cities by implementing DD estimation as a regression. Write down the equation of the regression to be estimated. You should include age and gender as controls and run separate regressions for whites and blacks. What are the regression DD estimates of the effect of the Mariel Boatlift on unemployment and (log) wages? Explain what type of standard errors you are using and why. Comment on the results.

d) What identification assumption needs to be satisfied for this DD analysis to be an accurate measure of the causal effect of the Mariel boatlift on labour market outcomes? Explain.

e) Implement a regression-based check of this identification assumption by modifying the unemployment and wage regressions in part c) to include city-specific time trends. Comment on the results.

f) In the summer of 1994, tens of thousands of Cubans boarded boats destined for Miami in an attempt to emigrate to the United States. To prevent this from happening, the Clinton administration interceded and ordered the Navy to divert the would-be immigrants to a base in Guantanamo Bay. Only a small fraction of Cubans ever reached Miami. This became known as the “Mariel Boatlift That Did Not Happen”. The dataset marielper2.dta on the course website contains data similar to marielper1.dta, but for the period from 1992 to 2002. Explain how you can use this non-event to check the validity of the DD identification assumption that you discussed in part d). Implement this check by running DD regressions like those estimated in part c) for this non-event. What do you conclude?

References:

Card, David (1990), “The Impact of the Mariel Boatlift on the Miami Labor Market”,

Industrial and Labor Relations Review, Vol. 43, No. 2., pp. 245-257

Question 2 (50 marks)

You are interested in studying whether an increase in mortgage credit by banks causes an increase in house prices. The dataset credit.dta on the course website contains data for counties in the United States for the period from 1995 to 2005 on the log change in the house price index and on three measures of credit growth: the log change in the number of loans, the log change in the loan volume and the log change in the loan-to-income ratio (defined as the amount of the loan divided by the income of the borrower).

a)   Suppose that you run the following regression of house price growth on credit growth:

𝑙𝑛𝑃𝑐,𝑡 − 𝑃𝑐,𝑡−1 = 𝛿(𝑙𝑛𝐿𝑐,𝑡 − 𝑙𝑛𝐿𝑐,𝑡−1)+ 𝛼𝑐 + 𝛾𝑡 + 𝑐,𝑡

where 𝑐 indexes counties and 𝑡 indexes years. 𝑃_𝑐,𝑡 is the county house price index, 𝐿_𝑐,𝑡 is one of the three measures of credit growth (number of loans, loan volume or loan- to-income ratio), 𝛼_𝑐 are county fixed effects and 𝛾_𝑡 are year fixed effects.

1. Explain why you would include county and year fixed effects in this regression.
2. Would this regression identify the causal effect of an increase in credit on house prices? Explain the possible types of bias in such a regression.

To overcome the identification challenges discussed in part a), Favara and Imbs (2015) adopt an instrumental variables (IV) strategy using regulatory changes to bank branching across states as an instrument for credit growth. In 1994, the US adopted legislation allowing banks to open branches across state borders without needing authorisation from state authorities. However, states still had some power to limit the entry of out-of-state branches. The dataset contains an index of restrictions to interstate branching (inter_bra). The index covers the period from 1994 to 2005 and takes values between 0 and 4, with high values referring to deregulated states.

The dataset contains data on credit growth for two types of institutions: commercial banks (indexed by _b) and independent mortgage companies (IMCs, indexed by _pl). Commercial banks use branches to collect deposits and originate loans, while IMCs rely on wholesale

funding and mortgage brokers. Only banks should respond to branching deregulation, as IMCs are unaffected by it.

In the analysis that follows, and in line with Favara and Imbs (2015), you should estimate all regressions by weighted least squares, with the weights given by the inverse of the number of counties per state (variable w1 in the dataset).

b)   What conditions does the deregulation index need to meet to be a valid instrument for credit growth?

c)    Run the first-stage regressions for the three measures of credit growth for commercial banks:

𝑙𝑛𝐿𝑐,𝑡 − 𝐿𝑐,𝑡−1 = 𝛽𝐷𝑠,𝑡−1 + 𝛼𝑐 + 𝛾𝑡 + 𝑐,𝑡

where 𝐷_{𝑠,𝑡−1} is the (lagged) deregulation index in state 𝑠 and the other variables are as described in part a). Explain what type of standard errors you are using and why.

Comment on the first-stage results.

d)   Estimate the same regressions as in part c) for lending by IMCs. How do your results compare with the ones for commercial banks? Explain how these results provide information on the validity of the identification assumptions discussed in part b).

e)   Run the reduced-form regression:

𝑙𝑛𝑃𝑐,𝑡 − 𝑃𝑐,𝑡−1 = 𝛽𝐷𝑠,𝑡−1 + 𝛼𝑐 + 𝛾𝑡 + 𝑐,𝑡

What do you conclude about the causal effect of branching deregulation on house prices?

f)    Estimate the IV regressions for the effect of the three measures of credit growth for commercial banks on house prices using deregulation as an instrument. What do you conclude about the causal effect of credit growth on house prices